Vol. I · 2026
RAGrets

No RAGrets, yet another AI security blog

More regrets


  • 4 min read

    Your RAG pipeline has no idea who is asking

    A multi-tenant retrieval setup tagged every chunk with a tenant ID and then never used it. How that happens, how to spot it in your own code, and what the fix looks like.

    rag · vector-db · multi-tenancy · access-control


  • 4 min read

    Ollama on 0.0.0.0, and other ways to give away your GPU

    Local model servers ship without authentication because they assume localhost. One environment variable, or one Docker port mapping, turns that into a public unauthenticated API.

    ollama · inference · networking · exposure